templates: Add Ollama multi-modal templates (#14868)

Templates for [local multi-modal
LLMs](https://llava-vl.github.io/llava-interactive/) using -
* Image summaries
* Multi-modal embeddings

---------

Co-authored-by: Erick Friis <erick@langchain.dev>

templates/rag-multi-modal-local/.gitignore

@@ -0,0 +1,2 @@
+docs/img_*.jpg
+chroma_db_multi_modal

templates/rag-multi-modal-local/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023 LangChain, Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

templates/rag-multi-modal-local/README.md

@@ -0,0 +1,122 @@
+
+# rag-multi-modal-local
+
+Visual search is a famililar application to many with iPhones or Android devices: use natural language to search across your photo collection. 
+  
+With the release of open source, multi-modal LLMs it's possible to build this kind of application for yourself and have it run on your personal laptop. 
+
+This template demonstrates how to perform visual search and question-answering over a collection of photos.
+ 
+Given a set of photos, it will use OpenCLIP embeddings to index them, retrieve photos relevant to user question, and use Ollama to run a local, open-source multi-modal LLM to answer questions about the retrieved photos.
+
+## Input
+
+Supply a set of photos in the `/docs` directory. 
+
+By default, this template has a toy collection of 3 food pictures.
+
+Example questions to ask can be:
+```
+What kind of soft serve did I have?
+```
+
+In practice, a larger corpus of images can be tested.
+
+To create an index of the images, run:
+```
+poetry install
+python ingest.py
+```
+
+## Storage
+
+This template will use [OpenCLIP](https://github.com/mlfoundations/open_clip) multi-modal embeddings to embed the images.
+
+You can select different embedding model options (see results [here](https://github.com/mlfoundations/open_clip/blob/main/docs/openclip_results.csv)).
+
+The first time you run the app, it will automatically download the multimodal embedding model.
+
+By default, LangChain will use an embedding model with moderate performance but lower memory requirments, `ViT-H-14`.
+
+You can choose alternative `OpenCLIPEmbeddings` models in `rag_chroma_multi_modal/ingest.py`:
+```
+vectorstore_mmembd = Chroma(
+    collection_name="multi-modal-rag",
+    persist_directory=str(re_vectorstore_path),
+    embedding_function=OpenCLIPEmbeddings(
+        model_name="ViT-H-14", checkpoint="laion2b_s32b_b79k"
+    ),
+)
+```
+
+## LLM
+
+This template will use [Ollama](https://python.langchain.com/docs/integrations/chat/ollama#multi-modal).
+
+Download the latest version of Ollama: https://ollama.ai/
+
+Pull the an open source multi-modal LLM: e.g., https://ollama.ai/library/bakllava
+
+```
+ollama pull bakllava
+```
+
+The app is by default configured for `bakllava`. But you can change this in `chain.py` and `ingest.py` for different downloaded models.
+
+## Usage
+
+To use this package, you should first have the LangChain CLI installed:
+
+```shell
+pip install -U langchain-cli
+```
+
+To create a new LangChain project and install this as the only package, you can do:
+
+```shell
+langchain app new my-app --package rag-chroma-multi-modal
+```
+
+If you want to add this to an existing project, you can just run:
+
+```shell
+langchain app add rag-chroma-multi-modal
+```
+
+And add the following code to your `server.py` file:
+```python
+from rag_chroma_multi_modal import chain as rag_chroma_multi_modal_chain
+
+add_routes(app, rag_chroma_multi_modal_chain, path="/rag-chroma-multi-modal")
+```
+
+(Optional) Let's now configure LangSmith. 
+LangSmith will help us trace, monitor and debug LangChain applications. 
+LangSmith is currently in private beta, you can sign up [here](https://smith.langchain.com/). 
+If you don't have access, you can skip this section
+
+```shell
+export LANGCHAIN_TRACING_V2=true
+export LANGCHAIN_API_KEY=<your-api-key>
+export LANGCHAIN_PROJECT=<your-project>  # if not specified, defaults to "default"
+```
+
+If you are inside this directory, then you can spin up a LangServe instance directly by:
+
+```shell
+langchain serve
+```
+
+This will start the FastAPI app with a server is running locally at 
+[http://localhost:8000](http://localhost:8000)
+
+We can see all templates at [http://127.0.0.1:8000/docs](http://127.0.0.1:8000/docs)
+We can access the playground at [http://127.0.0.1:8000/rag-chroma-multi-modal/playground](http://127.0.0.1:8000/rag-chroma-multi-modal/playground)  
+
+We can access the template from code with:
+
+```python
+from langserve.client import RemoteRunnable
+
+runnable = RemoteRunnable("http://localhost:8000/rag-chroma-multi-modal")
+```

templates/rag-multi-modal-local/ingest.py

@@ -0,0 +1,35 @@
+import os
+from pathlib import Path
+
+from langchain.vectorstores import Chroma
+from langchain_experimental.open_clip import OpenCLIPEmbeddings
+
+# Load images
+img_dump_path = Path(__file__).parent / "docs/"
+rel_img_dump_path = img_dump_path.relative_to(Path.cwd())
+image_uris = sorted(
+    [
+        os.path.join(rel_img_dump_path, image_name)
+        for image_name in os.listdir(rel_img_dump_path)
+        if image_name.endswith(".jpg")
+    ]
+)
+
+# Index
+vectorstore = Path(__file__).parent / "chroma_db_multi_modal"
+re_vectorstore_path = vectorstore.relative_to(Path.cwd())
+
+# Load embedding function
+print("Loading embedding function")
+embedding = OpenCLIPEmbeddings(model_name="ViT-H-14", checkpoint="laion2b_s32b_b79k")
+
+# Create chroma
+vectorstore_mmembd = Chroma(
+    collection_name="multi-modal-rag",
+    persist_directory=str(Path(__file__).parent / "chroma_db_multi_modal"),
+    embedding_function=embedding,
+)
+
+# Add images
+print("Embedding images")
+vectorstore_mmembd.add_images(uris=image_uris)

templates/rag-multi-modal-local/pyproject.toml

@@ -0,0 +1,38 @@
+[tool.poetry]
+name = "rag-multi-modal-local"
+version = "0.1.0"
+description = "Multi-modal RAG using Chroma"
+authors = [
+    "Lance Martin <lance@langchain.dev>",
+]
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = ">=3.8.1,<4.0"
+langchain = ">=0.0.351"
+openai = "<2"
+tiktoken = ">=0.5.1"
+chromadb = ">=0.4.14"
+open-clip-torch = ">=2.23.0"
+torch = ">=2.1.0"
+langchain-experimental = "^0.0.43"
+langchain-community = ">=0.0.4"
+
+[tool.poetry.group.dev.dependencies]
+langchain-cli = ">=0.0.15"
+
+[tool.langserve]
+export_module = "rag_multi_modal_local"
+export_attr = "chain"
+
+[tool.templates-hub]
+use-case = "rag"
+author = "LangChain"
+integrations = ["Ollama", "Chroma"]
+tags = ["multi-modal"]
+
+[build-system]
+requires = [
+    "poetry-core",
+]
+build-backend = "poetry.core.masonry.api"

templates/rag-multi-modal-local/rag_multi_modal_local.ipynb

@@ -0,0 +1,52 @@
+{
+ "cells": [
+  {
+   "attachments": {},
+   "cell_type": "markdown",
+   "id": "681a5d1e",
+   "metadata": {},
+   "source": [
+    "## Run Template\n",
+    "\n",
+    "In `server.py`, set -\n",
+    "```\n",
+    "add_routes(app, chain_rag_conv, path=\"/rag-multi-modal-local\")\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "d774be2a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langserve.client import RemoteRunnable\n",
+    "\n",
+    "rag_app = RemoteRunnable(\"http://localhost:8001/rag-multi-modal-local\")\n",
+    "rag_app.invoke(\" < keywords here > \")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.16"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

templates/rag-multi-modal-local/rag_multi_modal_local/__init__.py

@@ -0,0 +1,3 @@
+from rag_multi_modal_local.chain import chain
+
+__all__ = ["chain"]

templates/rag-multi-modal-local/rag_multi_modal_local/chain.py

@@ -0,0 +1,122 @@
+import base64
+import io
+from pathlib import Path
+
+from langchain.chat_models import ChatOllama
+from langchain.vectorstores import Chroma
+from langchain_core.documents import Document
+from langchain_core.messages import HumanMessage
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.pydantic_v1 import BaseModel
+from langchain_core.runnables import RunnableLambda, RunnablePassthrough
+from langchain_experimental.open_clip import OpenCLIPEmbeddings
+from PIL import Image
+
+
+def resize_base64_image(base64_string, size=(128, 128)):
+    """
+    Resize an image encoded as a Base64 string.
+
+    :param base64_string: A Base64 encoded string of the image to be resized.
+    :param size: A tuple representing the new size (width, height) for the image.
+    :return: A Base64 encoded string of the resized image.
+    """
+    img_data = base64.b64decode(base64_string)
+    img = Image.open(io.BytesIO(img_data))
+    resized_img = img.resize(size, Image.LANCZOS)
+    buffered = io.BytesIO()
+    resized_img.save(buffered, format=img.format)
+    return base64.b64encode(buffered.getvalue()).decode("utf-8")
+
+
+def get_resized_images(docs):
+    """
+    Resize images from base64-encoded strings.
+
+    :param docs: A list of base64-encoded image to be resized.
+    :return: Dict containing a list of resized base64-encoded strings.
+    """
+    b64_images = []
+    for doc in docs:
+        if isinstance(doc, Document):
+            doc = doc.page_content
+        # Optional: re-size image
+        # resized_image = resize_base64_image(doc, size=(1280, 720))
+        b64_images.append(doc)
+    return {"images": b64_images}
+
+
+def img_prompt_func(data_dict, num_images=1):
+    """
+    GPT-4V prompt for image analysis.
+
+    :param data_dict: A dict with images and a user-provided question.
+    :param num_images: Number of images to include in the prompt.
+    :return: A list containing message objects for each image and the text prompt.
+    """
+    messages = []
+    if data_dict["context"]["images"]:
+        for image in data_dict["context"]["images"][:num_images]:
+            image_message = {
+                "type": "image_url",
+                "image_url": f"data:image/jpeg;base64,{image}",
+            }
+            messages.append(image_message)
+    text_message = {
+        "type": "text",
+        "text": (
+            "You are a helpful assistant that gives a description of food pictures.\n"
+            "Give a detailed summary of the image.\n"
+            "Give reccomendations for similar foods to try.\n"
+        ),
+    }
+    messages.append(text_message)
+    return [HumanMessage(content=messages)]
+
+
+def multi_modal_rag_chain(retriever):
+    """
+    Multi-modal RAG chain,
+
+    :param retriever: A function that retrieves the necessary context for the model.
+    :return: A chain of functions representing the multi-modal RAG process.
+    """
+    # Initialize the multi-modal Large Language Model with specific parameters
+    model = ChatOllama(model="bakllava", temperature=0)
+
+    # Define the RAG pipeline
+    chain = (
+        {
+            "context": retriever | RunnableLambda(get_resized_images),
+            "question": RunnablePassthrough(),
+        }
+        | RunnableLambda(img_prompt_func)
+        | model
+        | StrOutputParser()
+    )
+
+    return chain
+
+
+# Load chroma
+vectorstore_mmembd = Chroma(
+    collection_name="multi-modal-rag",
+    persist_directory=str(Path(__file__).parent.parent / "chroma_db_multi_modal"),
+    embedding_function=OpenCLIPEmbeddings(
+        model_name="ViT-H-14", checkpoint="laion2b_s32b_b79k"
+    ),
+)
+
+# Make retriever
+retriever_mmembd = vectorstore_mmembd.as_retriever()
+
+# Create RAG chain
+chain = multi_modal_rag_chain(retriever_mmembd)
+
+
+# Add typing for input
+class Question(BaseModel):
+    __root__: str
+
+
+chain = chain.with_types(input_type=Question)